The MPEG-1 standard was developed in response to the industry need of implementing an efficient way of storing and retrieving video information on storage media of the digital type, as for example CD-ROMs. Of course, the MPEG-1 standard is also a powerful tool for efficiently storing data on similar media such as DATs, Winchester disks, optical disks and ISDN and LAN networks. A more efficient version of the standard called MPEG has been developed in support of bitrate requisites in the field of digital video transmission applications. The standard has been generally accepted for digital TV systems, for compressing TV-resolution pictures entirely interlaced and up to a bitrate of about 15 Mbps. A special version of the MPEG standard is expected to be used in future generation HDVT systems.
The MPEG standard incorporates and uses important algorithms and criteria defined by previous international standards, such as, for example, the CCITT motion vectors determination algorithm H.261 and the ISO 10918 standard of the ISO JPEG Committee for the coding of still pictures. A definition of the MPEG standard (1 and 2), and a thorough description of the different application techniques and the relative coding and decoding systems of the data relative compressed video pictures according to the MPEG standards are described in a wealth of articles and publications on the subject, including the following:
Draft International ISO/IEC DIS 13818-2 "Information technology-Generic coding of moving pictures and associated audio information". PA0 "MPEG coding and transport system" by Leonardo Chiariglione, Digital Television Broadcasting-Proceedings. PA0 "The MPEG video compression algorithm" by Didier J. Le Gall, Signal Processing Image Communication, Elsevier Science Publishers B. V., Vol. 4, No. 2, April 1992. PA0 Digest No. 1995/012, Electronics Division, Institution of Electrical Engineers-London, Colloquium on: "MPEG-2--what it is and what it isn'2". PA0 "An Overview of the MPEG-2 Compression Algorithm" Technical Note released by SGS-THOMSON MICROELECTRONICS (An 529/0294). PA0 Datasheet "STi3500A" Datasheet of SGS-THOMSON MICROELECTRONICS. PA0 "STi3520A--Advanced Information for an MPEG Audio/MPEG-2 Video Integrated Decoder" (June 1995). PA0 A "Bit buffer": that is a buffer for compressed data that the MPEG-2 standard fixes at 1.75 Mbits plus an extra amount, for example, of 983.040 bits, in consideration of a nonideal process of decompression being actually carried out. PA0 A first "I-frame buffer" for the decompressed Intra-picture or briefly I-picture, in a 4:2:0 format. PA0 A second "P-frame buffer" for the decompressed Predicted-picture or briefly P-picture, in a 4:2:0 format. PA0 A third "B-frame buffer" for the decompressed Bidirectionally Predicted Picture or briefly B-picture, in a 4:2:0 format, eventually optimized so to require a reduced amount of memory, which is 0.7407 or 0.6111 of a frame, respectively, in a PAL or NTSC system. PA0 1.835.008+835.584+4.976.640+4.976.640=12.623.872 bits PA0 . . A1 A2 A3 A4 A5 A6 A7 A8 A9 A10 A11 A12 A13 A14 A5 A16 . . .
According to a typical architecture of an MPEG-2 decoder, such as that shown in FIG. 3 of the publication No. STi3520A relative to an MPEG Audio/MPEG-2 Video integrated decoder marketed by SGS-THOMSON MICROELECTRONICS, herein reproduced as FIG. 1, there exist well-defined requirements of video memory. These requirements, that is, the capacity of an external DRAM memory that, for a PAL and NTSC application can support 16 Mbits PAL video signals, can be estimated as follows.
Considering that both the MPEG-2 video decoder and the MPEG audio decoder access a unique external DRAM memory of 16 Mbits through a common interface, the audio decoder may require access to only 131.072 bits thereby leaving the remaining 16.646.144 bits available for satisfying the requisites of the MPEG-2 video decoder. The video memory can be configured as follows:
According to the actual MPEG-2 standard technique, and despite dealing with an I, P or B-picture, dependently on the type of video standard, each "frame buffer" in the 4:2:0 format may occupy an amount of memory given by the following table.
______________________________________
720 .times. 576 .times. 8 for the luma 3.317.760 bits = 4.976.640 (luminance) (Y) bits 360 .times. 288 .times. 8 for the U chroma 829.440 bits (chrominance U) 360 .times. 288 .times. 8 for the V chroma 829.440 bits (chrominance V) NTSC 720 .times. 480 .times. 8 for the luma 2.764.800 bits = 4.147.200 (luminance) (Y) bits 360 .times. 240 .times. 8 for the U chroma 691.200 bits (chrominance U) 360 .times. 240 .times. 8 for the V chroma 691.200 bits (chrominance V) ______________________________________
Therefore, in a PAL system, which representing the most burdensome case may serve as a reference example, the actual total amount of memory required will be given by: 1.835.008+835.584+4.976.640+4.976.640+(4.976.640*0.7407)=16.310.070 bits. This calculation considers a 0.7407 optimization of the B-picture frame buffer.
A further optimization may include undertaking the decompression of the B-picture without resorting to a storage step in the external RAM by carrying out an equivalent function internally in the integrated decoder device by a dedicated circuit block functionally placed upstream of the Display Unit.
Considering this further optimization, the video RAM requirement drops to:
where the B-buffer is realized within the same chip containing the "core" of the decoder being required to convert the scanning of each 8*8 block, defined in the MPEG-2 compressed data stream, in that of each row of the picture (field or frame) required by the video display process of the picture itself. Such conversion macrocell is commonly called "MACROBLOCK TO RASTER SCAN CONVERTER".
In European patent application No. 96830106.9, filed on 11 Mar. 1996 (corresponding to U.S. application Ser. No. 08/758,805 filed Dec. 3, 1996), in the name of the same assignee, a method and relative device were described that allow a significant reduction of the above cited video memory requirement down to about 8 Mbits.
The idea behind the invention described and claimed in the above identified prior patent application is the recognition that the amount of memory required by the MPEG decoding process as stated in the above calculations can be remarkably reduced when allowing for a recompression of the pictures used as a reference for the prediction (I-picture and P-picture for the case of the standards MPEG-1 and MPEG-2), after the MPEG decompression and before they are temporarily stored in the external video memory and their decompression when they are read from the external memory. On the other hand, digital TV receivers commonly employ an integrated system that is generally known as "Improved Quality Television processor" (IQTV) or "Smart Quality Television processor" (SQTV), which requires storing of whole pictures or parts of them to process them during the decoding phase. These integrated systems are increasingly employed in video receivers to improve image quality without requiring changes in the relevant transmission standards (PAL, NTSC, SECAM).
In practice, luminance (Y) and chrominance-U and chrominance-V (UV) analog components of the video signal are digitized through analog-to-digital converters and the digital data obtained, concerning the two fields into which the video signal is generally divided, are stored in a dedicated memory. This allows, first, a conversion from 50 or 60 Hz to 100 or 120 Hz of the interlaced scanning frequency (in practice the so-called "refresh-rate") so to make flickering imperceptible.
Moreover, digital values relevant to each pixel of each line forming each field so stored in the memory enable implementing appropriate algorithms for reducing Gaussian and spike or pulse noise. Through memory duplication it is also possible to implement a compensation (elimination) loop for blurring of objects moving fast on a background by employing an algorithm that contemplates the calculation of average values among the values of pixels relating to successive pictures. This eliminates non-correlated contents of the video signal and correction algorithms based on motion detection. The field memory or memories are normally DRAM devices which, according to the present limits of manufacturing technologies, are external to the integrated device containing the processing circuits (generally called SQTV.sub.-- IC or IQTV.sub.-- IC).
A common video receiver includes a tuning section that selects the channel and the relevant frequency band of the signal transmitted via cable, satellite, or is broadcast. The tuner converts the received signal to an intermediate video frequency (for example 38.9 MHz according to the Italian standard) containing luminance and chrominance data multiplexed in frequency. A demodulation block processes the signal converted at the intermediate frequency producing a base band composite video signal which, in turn, is selected by a selector of the input signal source and which feeds the standard decoder (PAL, NTSC, SECAM) carrying out the separation of luminance and chrominance components.
Analog signals relevant to luminance (Y) and chrominance (UV) are thus converted through appropriate analog-to-digital converters to produce a luminance (Y) digital data stream and a chrominance (UV) digital data stream which are fed to the IQTV.sub.-- IC (or SQTV.sub.-- IC) device). Each FIELD MEMORY may store a picture field in the 4:2:2 format or in the 4:1:1 format, thus occupying the following memory space.
______________________________________
720 .times. 288 .times. 8 for luma 1,658,880 bits = 3,317,760 (luminance) (Y) bits 360 .times. 288 .times. 8 for U chroma 829,440 bits (chrominance U) 360 .times. 288 .times. 8 for chroma 829,440 bits (chrominance V) NTSC 720 .times. 240 .times. 8 for luma 1,382,400 bits = 2,764,800 (luminance) (Y) bits 360 .times. 240 .times. 8 for U chroma 691,200 bits (chrominance U) 360 .times. 240 .times. 8 for chroma 691,200 bits (chrominance V) ______________________________________
In the case of a PAL transmission system which, being the more critical standard, can be taken as a reference example, the total requisite when employing two field memories becomes 3,317,760*2=6,635,520 bits.
Even for this type of application, there is the need or usefulness to reduce this memory requisite of a SQTV.sub.-- IC (or IQTV.sub.-- IC) without causing a perceivable visual deterioration of images. This would be desirable to save costs and/or create more suitable opportunities for the integration of the field memory in the same IQTV.sub.-- IC (or SQTV.sub.-- IC) chip through improvements of manufacturing processes.
In another preceding patent application of the same assignee: EP-96830134.1, of Mar. 20, 1996 (corresponding to U.S. application Ser. No. 08/810,029, filed Mar. 4, 1997), an IQTV-IC system is disclosed which implements a compressing and coding according to an adaptive differential pulse code modulation scheme (ADPCM) of the digital data relevant to a field in a 4:2:2 or in a 4:1:1 format of a picture to be stored in the relevant field memory at a 50 or 60 Hz standard frequency. The application further disclosures an ADPCM decoding and decompressing of the data read from the field memory at a multiple frequency (generally double) of the frequency at which the data are written in the field memory.
The ADPCM compression mechanism takes advantage of the correlation existing between adjacent pixel values of a picture to reduce the number of bits required for a binary representation of the picture itself. In fact, it is possible to approximate the pixel values by suitably combining only the value of pixels adjacent thereto (therefore, not using (storing) the actual pixel value), that is, by implementing a so-called "prediction" of the pixel itself.
Therefore, it is possible to reduce the quantity of binary figures required for the digital representation of a picture by defining the prediction mechanism and therefore duly coding (and storing) the prediction error only instead of each pixel. The more accurate the prediction of the pixel value, the lower the entropy of the prediction error will be, or the lower the number of bits required to code it will be.
The method described in the above mentioned prior patent application permits in practice to more than halve the RAM memory requirement for the storing of the field or fields, and with an almost neglectible loss in the quality of reconstructed images since artifacts are mainly distributed at high frequencies. In both the systems described in the prior patent applications, a significative feature was the presence of an ADPCM coder that recompressed the MPEG decoded digital video data, before storing it in the respective buffer to reduce the video memory requisite.
The variance estimator is that portion of an ADPCM coder that performs an estimation of the power of the video signal contained in the portion of the data stream being compressed. In fact, according to an ADPCM technique, the compression "engine" is represented by a parametric quantizer. The set of parameters or coefficients used by the parametric quantizer depend from the power of the video block and each of the sets of parameters or coefficients correspond to a precise address of a dedicated read only memory (look-up table) containing them. These sets of parameters or coefficients are precalculated and loaded in a dedicated ROM or PAL device. Therefore, the variance estimator provides the address of the ROM or PAL containing the parameters which permit the quantizer to compress the preevaluated block of video data with an error that is smaller the more precisely the appropriate set of precalculated parameters are specific.
A way of ADPCM compressing data relative to a digital video signal as described in the above mentioned prior patent applications is based on the following criteria. By considering an 8*8 (8 columns and 8 rows) block of pels, let us calculate the differences of moduli of adjacent pels of the same row; and to calculate also the differences among adjacent pels belonging to a first column of the 8*8 block. In the following table the considered video block is shown in the left-hand side, while the block in the right-hand side shows the differences of moduli among adjacent pels:
__________________________________________________________________________ A1 A2 A3 A4 A5 A6 A7 A8 0 A(1-2) A(2-3) A(3-4) A(4-5) A(5-6) A(6-7) A(7-8) B1 B2 B3 B4 B5 B6 B7 B8 B(1-2) B1-A1 B(2-3) B(3-4) B(4-5) B(5-6) B(6-7) B(7-8) C1 C2 C3 C4 C5 C6 C7 C8 C(1-2) C1-B1 C(2-3) C(3-4) C(4-5) C(5-6) C(6-7) C(7-8) D1 D2 D3 D4 D5 D6 D7 D8 D(1-2) D1-C1 D(2-3) D(3-4) D(4-5) D(5-6) D(6-7) D(7-8) E1 E2 E3 E4 E5 E6 E7 E8 E(1-2) E1-D1 E(2-3) E(3-4) E(4-5) E(5-6) E(6-7) E(7-8) F1 F2 F3 F4 F5 F6 F7 F8 F(1-2) F1-E1 F(2-3) F(3-4) F(4-5) F(5-6) F(6-7) F(7-8) G1 G2 G3 G4 G5 G6 G7 G8 G(1-2) G1-F1 G(2-3) G(3-4) G(4-5) G(5-6) G(6-7) G(7-8) H1 H2 H3 H4 H5 H6 H7 H8 H(1-2) H1-G1 H(2-3) H(3-4) H(4-5) H(5-6) H(6-7) H(7-8) __________________________________________________________________________
Let us sum the differences, divide the result by 64(8*8), and finally multiply the result by a correction coefficient. The result is the address of the ROM containing precalculated sets of coefficients of quantization. The reading of these coefficients initializes a quantizer that, according to an ADPCM technique, compresses the video block depicted in the left-hand side portion of the above table according to the criteria shown in the right-hand side block of the table.
In the cited prior patent application EP-96830367.7, filed on Jun. 28, 1996 (corresponding to U.S. application Ser. No. 08/880,260), a method and relative device were described which provided for a further reduction of the video memory requisite. This was accomplished through the performance of a subsampling operation of the data before recompressing them with an ADPCM technique, at least for the I and P pictures, after MPEG-2 decompression, and before storing the ADPCM recompressed data in respective buffers of the video memory. Subsequently, during reconstruction of the blocks of pels to be sent to a "MACROBLOCK-SCAN TO RASTER-SCAN" conversion unit, the decoded and decompressed data were oversampled congruently with the subsampling factor that was used for subsampling them before ADPCM recompression.